passive bistatic sar imaging and interferometry by ...jmfriedt.free.fr/igarss2019.pdfdigital storage...
TRANSCRIPT
Abstract—In this study, we used the digital television signal
broadcasted by geostationary satellite for passive bistatic synthetic
aperture radar (SAR) imaging and interferometry applications. A
prototype system was designed and fabricated with commercial-
off-the-shelf (COTS) components. Specifically, three low noise
block downconverters (LNBs) were synchronized for a long-time
coherent measurement. Moreover, multiple TV channels were
combined to achieve a wider frequency bandwidth to improve the
range resolution. The potential applications of this ground-based
radar system are SAR imaging, displacement estimation, and
digital elevation model (DEM) generation. Hardware design and
data processing algorithms are described. Experimental results
are presented to validate the performance of the designed system
and the proposed methods.
Index Terms—Passive bistatic radar (PBR), Satellite digital TV
signal, Synthetic aperture radar (SAR), SAR interferometry.
I. INTRODUCTION
Recently, passive bistatic radar that uses existing broadcasting
signals, communication signals, and navigation signals has
drawn a lot of interests [1-2]. Compared to other illuminators of
opportunity, such as WiFi and terrestrial TV signal, PBR using
satellite digital TV signal has several advantages, including: 1)
one TV channel can provide a frequency bandwidth larger than
30 MHz; if multiple TV channels are combined, a bandwidth
larger than 150 MHz and thus a sub-meter range resolution can
be obtained; 2) since its working frequency is at Ku band, a
high-resolution change detection capability via interferometry
technique can be achieved, like that in ground-based SAR [3-
4]; 3) a larger illumination area and a clearer reference signal
increase the usability and performance of the PBR system. For
example, digital TV broadcasting with Broadcasting Satellite
(BS) on the geostationary orbit of 110 degrees East longitude
(BS 110°E) can provide services for the whole area of Japan.
However, there are also some difficulties when using the
satellite TV signal for PBR applications, including: 1) the
power density of the satellite TV signal on the Earth surface is
normally much lower than some other signals used for PBR
applications, such as FM and DAB, resulting in a lower signal
to noise ratio (SNR) and a shorter detection range; 2) if some
commercial-off-the-shelf (COTS) components are used to
reduce the system cost, some modifications must be conducted
to synchronize both frequency and phase between the reference
channel and the surveillance channels for long-time coherent
measurement; 3) the waveform of the digital TV signal is not
designed for radar applications, its signal structure, including
spectrum properties, pilot signals, and guard intervals, will
introduce undesired artifacts in the range direction when the
classical cross-correlation processing method is used; 4) when
multiple frequency bands were combined to increase the range
resolution, frequency gaps among different TV channels cause
high-level artifacts in the range direction [5].
To solve these problems, we proposed a hardware design
method and several processing algorithms in this paper. Firstly,
we describe an LNB synchronization method to increase the
integration time for SNR improvement. Secondly, we use the
inverse filter technique to reduce the influence of the TV signal
structure. Thirdly, we apply the super spatially variant
apodization (Super-SVA) method to fill the frequency gaps
among multiple TV channels to suppress the artifacts. Finally,
based on a PBR prototype system using COTS components,
SAR imaging and displacement estimation experiments are
carried out to validate the proposed methods.
II. DESIGNED SYSTEM OVERVIEW
A. Adopted ISDB-S signal
Fig. 1. The spectrum of the satellite TV signal in Japan.
The spectrum of the used satellite TV signal is shown in Fig.
1, where the local oscillator frequency of LNB is about 10.601
GHz. BS and communication satellite (CS) signals are
broadcast by different satellites that are on the geostationary
orbit of 110 degrees East longitude (110°E). Each TV channel
has a frequency bandwidth of 34.5 MHz. If twelve channels of
BS or CS are combined, a 475.5 MHz frequency bandwidth can
be achieved, resulting in a maximal bistatic range resolution of
0.32 m. We note that, in this study, only CS signal is used for
simplicity. BS signal transmitted via both left-hand and right-
PASSIVE BISTATIC SAR IMAGING AND INTERFEROMETRY BY SATELLITE
DIGITAL TV SIGNAL
1Weike Feng, 2Jean-Michel Friedt, 3Giovanni Nico, 2Gilles Martin, 4Motoyuki Sato
1Graduate School of Environmental Studies, Tohoku University, Sendai, Japan
2FEMTO-ST, Time & Frequency department, Besancon, France 3Consiglio Nazionale delle Ricerche, Istituto per le Applicazioni del Calcolo, Bari, Italy
4Center for Northeast Asian Studies, Tohoku University, Sendai, Japan
hand circular polarizations will be used for multi-polarization
applications in the near future.
B. Designed radar system
The designed radar system is shown in Fig. 2. It has three
parallel receiving channels, one reference channel and two
surveillance channels. For reference channel, a 45-cm parabolic
antenna is used, while only the feed-horns of parabolic antennas
are used for the surveillance channels to achieve a wider beam-
width. Three LNBs with a local frequency of 10.601 GHz are
synchronized by an oscillator controller. To further amplify the
received signal, three COTS BS/CS boosters were used, whose
working frequency is from 10 to 2600 MHz and gain is from 26
to 34 dB adjustable. Surveillance antennas are mounted on a
positioner to generate a synthetic aperture with a fixed step in
the horizontal direction. The signal flow will be sampled by a
digital storage oscilloscope with a sampling frequency of 10
GHz. Finally, the acquired raw data are transferred to a local
PC for off-line signal processing, consisting in frequency gaps
filling, SAR imaging, and interferometric processing.
Fig. 2. The designed PBR system using satellite TV signal.
C. LNB synchronization
Since COTS LNBs have different local oscillators, their
phase differences will reduce the coherence of the three
receiving channels. When the data are sampled in a short time,
the phase difference is approximately caused by their local
frequencies only. Thus, a simple phase difference estimation
method can be used [5], consisting in: 1) Divide the signal into
U short-length (T) subsets, within which the phase difference is
assumed to be constant, i.e., f1t and f2t, where t=[0, T, ... , (U-
1)T], f1 and f2 are the frequency differences between the
surveillance LNBs and the reference LNB; 2) Do cross-
correlation between the co-located reference and surveillance
antennas to obtain the phase of peaks in each subset, which
gives e−j2πf1t and e−j2πf2t; 3) Finally, do Fourier transform of the
peak phases, providing us with f1 and f2.
However, since SAR imaging and interferometric processes
are desired and a longer integration time can increase the SNR
and increase the detection distance, coarse phase correction is
not enough. Thus, an advanced LNB synchronization approach
is further proposed, whose setup is shown in Fig. 3. Based on
the unpublished NXP application note "X-tal driver for using
multiple TFF10xx to the same X-tal as reference" describing
the mechanism for locking multiple TFF10xx PLLs on the same
reference as desired, a dedicated Pierce oscillator was built
around a 25 MHz resonator and a 74HC04 inverter. The output
of the Pierce oscillator feeds another 74HC04 used to generate
three sets of in-phase and phase-opposition (180-degree phase
shift) signals feeding the PLL inputs. Care must be taken on the
one hand to tune the output voltage amplitude to acceptable
levels (500 mVpp) using a voltage divider bridge, and on the
other hand to include a DC-blocking capacitor (5.6 nF) between
the oscillator output and the PLL to avoid DC-current leakage
from the PLL, which would prevent PLL locking. Under the
proper conditions, frequency/phase coherence was observed to
be stable over tens of minutes, as needed for SAR measurement.
Fig. 3. LNB synchronization setup.
III. INVERSE FILTER AND FREQUENCY GAP FILLING
A. Cross-correlation and inverse filter
Normally, a cross-correlation method is used to estimate the
delay of the target, given by
0( ) ( ) ( )
intTa
a sur refs t s t dt (1)
which can be implemented by Fast Fourier Transform, giving
( ) [ ( ) ( ) ]H H a
a sur ref
s f s f s f (2)
where sref(t) is reference signal, ssur(t) is surveillance signal, a =
1 or 2 denotes the first or the second surveillance channel, (·)∗
denotes the complex conjugate, χ(τ) is the range-compression
result, τ is the expected time delay of the target, Tint is the
integration time, Φ is the Fourier transform matrix, denotes
element-wise multiplication, sref(f) and ssur(f) are the frequency
domain signals of the reference and surveillance channels.
However, since the waveform of satellite TV signal is not
designed for radar applications, the cross-correlation processing
will introduce high-level artifacts in the estimation. A simple
but effective method to solve this problem is to use an inverse
filter to avoid the influence of the signal structure of the TV
signal, which can be expressed as 2( ) [ ( ) ( ) / | ( ) | ]H H a
a a sur ref ref
's f s f s f s f (3)
B. Frequency gap filling
To improve the range resolution, multiple TV channels can
be combined to obtain a wider bandwidth. However, different
TV channels are separated by frequency gaps to guarantee a
high-quality signal transmission, introducing artifacts into the
range compression result. Several spectral gaps filling methods
can be used, such as auto-regressive (AR) based method, Super-
SVA based method, and sparsity and low-rank property based
methods [5]. In this paper, Super-SVA method is used for its
simplicity and effectiveness. Based on the SVA algorithm, the
range compression result of the n-th (n = 1, 2,...,12) TV channel
is given by
Osc
illo
sco
pe
Ant. Ref
Ant. Surv-1
Co
ntr
oll
er a
nd
off
-
lin
e si
gn
al p
roce
sso
r
Position
controller
Ant. Surv-2
Pos
itio
ner
LNB oscillator controller
LO1LNB1
LO2LNB2
LO3LNB3 Booster
Booster
Booster
Target
, ,[ ( ) ( )]SVA H
a n n a n 's f W f (4)
where ( )W f is a nonlinear “cosine on pedestal” weighting
function that can be expressed as
( ) 1 2 cos(2 / ), 0 0.5W m m M (5)
Then, the frequency gaps can be filled by Super-SVA via the
following steps:
1) Perform Fourier transform of ,
SVA
a n to get , ( )SVA
a n f ;
2) Apply an inverse weighting vector ( )invW f to obtain
, ,( ) ( ) / ( )S SVA SVA
a n a n invW s f f f ;
3) Replace the center portion of the extrapolated signal
, ( )S SVA
a n
s f with the original signal
, ( )a n
's f .
According to this procedure, the gapped signal can be
reconstructed for each TV channel. Since two adjacent channels
have the same gapped frequencies, a simple average process
can be used to obtain the final estimation. Finally, the range
compression can be obtained by ( )S SVA H S SVA
a a
s f .
IV. SAR IMAGING AND INTERFEROMETRY
Fig. 4. SAR imaging geometry.
The imaging geometry of the designed radar system is shown
in Fig. 4, where the reference antenna is located at (0, 0) and the
surveillance antenna is scanned along a rail from ( / 2, )a
surL y
to ( / 2, )a
surL y with L as the synthetic aperture length. The
elevation angle of the satellite is 35.3sat in Sendai (Japan)
area. For the o-th (o=1, 2,..., O) surveillance antenna position,
the range compression can be conducted by applying Fourier
transform to the gap filled signal, giving , , ( )S SVA H S SVA
a o a o
s f .
Then, the SAR image of the target can be obtained by applying
the back projection (BP) algorithm, giving
, 1 21[ (( ) )],
O S SVA a
a ooa x R R oy
(6)
where ( , )a x y is the estimated amplitude of the point at (x, y),
1R is the path difference between the TV satellite to the target
and to the reference antenna, 2 2
2 ( )= ( ) ( )a a
o surR o x x y y is
the distance from the target to the o-th antenna position of the
a-th surveillance antenna, and , 1 2[ ( )]S SVA a
a o R R o can be
calculated by the linear interpolation process. Since the satellite
is far from the experiment position, i.e., |yT| is much bigger than
x and y, 1R can be approximately calculated by
2 2
1 ( )T TR x y y y y (7)
The designed passive radar system can acquire a time series
of 2D SAR images of the scene with a temporal baseline of few
minutes as the used geostationary satellite is continuously
illuminating the area of interest. This allows many interesting
interferometric SAR applications. For each pair of SAR images,
an interferometric phase image can be generated by
- ( , ) ( ,a ) (r )},g{ mater slD I aven x y x y x y
(8)
As an application, the displacement of a target can be
estimated by the relationship - ( , ) / 4D InD x y , where λ is
the wavelength, which is 24 mm for the CS signal. Besides, we
note that, since two surveillance antennas can be used, the
developed system can also generate a high resolution Digital
Elevation Model (DEM) of the imaging scene by properly
setting a spatial baseline between the two surveillance antennas.
DEM generation experiments are on-going and will be further
researched.
V. EXPERIMENT RESULTS
In this section, some experimental results will be shown to
validate the designed radar system and the proposed processing
methodologies. At the beginning, a time delay estimation
experiment using the satellite TV signal was conducted, as
shown in Fig. 5, where two parabolic antennas with
synchronized LNBs were faced toward the CS 110°E satellite.
Twelve TV channels were combined and inverse filter is used
for range compression. The estimation results are shown in Fig.
6, where two different integration times are assessed. As we can
see, by increasing the integration time from 1 μs to 100 μs, the
SNR can be increased about 10 dB, which proves that the LNBs
are well synchronized. Otherwise, the SNR will not be
increased but decreased because the two receiving channels are
not coherent [5].
Fig. 5. Time delay estimation experiment setup.
Fig. 6. Delay estimation results with different integration time.
However, as indicated by the black circle in Fig. 6, high-level
periodical artifacts will be generated by the frequency gaps. By
applying Super-SVA based frequency gap filling method, these
artifacts can be well suppressed, as shown in Fig. 7, where the
0 0( , )x y
0 0( , )
,( )a
o surx y
y
x
1R
2R
sat
0x
0y
0,( )Ty
integration time is 100 μs. Furthermore, in order to check the
coherence of two channels, 250 measurements were conducted
with a period of 5 minutes, and the peak phase of the result was
extracted, as shown in Fig. 8. It is observed that, the peak phase
difference has a mean value of -0.16 degree and a standard
deviation of 2.50 degree for 250 measurements.
Fig. 7. Delay estimation result obtained by filling frequency gaps.
Fig. 8. Phase stability evaluation for 250 measurements.
Based on the aforementioned methodologies, passive bistatic
SAR imaging experiments were carried out, as shown in Fig. 9,
where the surveillance antenna is sled along a rail to detect a
metallic plate. The moving step was 5 mm and the total sledding
length was 1.2 m. The range compression results for each
surveillance antenna position and the SAR image obtained by
the BP algorithm are shown in Fig. 10. As we can see, the
metallic plate can be well detected and focused.
Fig. 9. SAR imaging of a stationary metalli plate.
Fig. 10. Top: Range compression results. Bottom: SAR image.
At last, the displacement of the metallic plate, which was
moved on a rail from 1 mm to 15 mm with a 1-mm step, was
estimated. The acquisition time of each SAR image was about
45 seconds. The estimation results are shown in Fig. 11. It can
be seen that the displacement of the metallic plate can be
accurately estimated. The root mean square error (RMSE) of
the estimation is about 0.27 mm.
Fig. 11. Displacement estimation result.
VI. CONCLUSION
In this study, we developed a passive bistatic radar system
using geostationary satellite TV signal with commercial-off-
the-shelf hardware. LNB synchronization was used to increase
the coherence for the long-time measurement and thus increase
the SNR and the detection range. High range resolution was
obtained by combing multiple TV channels. The artifacts
caused by frequency gaps among different TV channels were
reduced by means of a super-SVA based method. High azimuth
resolution was achieved by moving the surveillance antenna on
a rail to generate a synthetic aperture. Focused SAR images of
targets were obtained by the back projection algorithm with
reasonable approximations. Displacement of the target in the
scene was estimated by the differential interferometric process.
Compared to current GB-SAR system used for displacement
estimation, the developed system is cheaper as it is does not
need a transmitting unit. Multiple-direction displacement
estimations can be obtained with multiple receivers using
different bistatic geometries. Further work will be devoted to
the generation of a DEM of the study area using an additional
surveillance antenna. Multi-polarization capacity provided by
the BS TV signal is also worthy to be studied. The potential
applications of the developed system includes the continuous
monitoring of landslides, buildings and dams.
ACKNOWLEDGEMENTS
This work was supported by JSPS Grant-in-Aid for Scientific
Research (A) 26249058. The passive radar investigation is
partly supported by the French Centre National de la Recherche
Scientifique (CNRS) PEPS grant.
REFERENCES
[1] H. D. Griffiths and C. J. Baker, “An introduction to passive radar,” Artech
House, 2017.
[2] D. Gromek, K. Kulpa, and P. Samczynski, “Experimental results of
passive SAR imaging using DVB-T illuminators of opportunity,” IEEE
Geosci. Remote Sens. Lett., vol. 13, no. 8, pp. 1124-1128, Aug. 2016.
[3] G. Nico, G. Cifarelli, G. Miccoli, F. Soccodato, W. Feng, M. Sato, S.
Miliziano, and M. Marini, “Measurement of pier deformation patterns by
ground-based SAR interferometry: Application to a bollard pull trial,”
IEEE J. Ocean. Eng., vol.43, no. 4, pp. 822-829, 2018.
[4] W. Feng, G. Nico, and M. Sato, “GB-SAR interferometry based on
dimension-reduced compressive sensing and multiple measurement
vectors model,” IEEE Geosci. Remote Sens. Lett., vol. 16, no. 1, pp. 70-
74, 2019.
[5] W. Feng, J. M. Friedt, and M. Sato, “Passive radar imaging by filling gaps
between ISDB digital TV channels,” IEEE J. Sel. Topics Appl. Earth
Observ. Remote Sens., in press.